Fix a race in tcpdns close with uv_close on timer

stop timers before closing

netmgr: tcpdns_close needs to be asynchronous, it manipulates sock->timer

lib/isc/netmgr/netmgr-int.h

@@ -122,6 +122,7 @@ typedef enum isc__netievent_type {
 	netievent_udpstoplisten,
 	netievent_tcpstoplisten,
 	netievent_tcpclose,
+	netievent_tcpdnsclose,
 	netievent_prio = 0xff,	/* event type values higher than this
 				 * will be treated as high-priority
 				 * events, which can be processed
@@ -194,6 +195,7 @@ typedef isc__netievent__socket_t isc__netievent_udpstoplisten_t;
 typedef isc__netievent__socket_t isc__netievent_tcpstoplisten_t;
 typedef isc__netievent__socket_t isc__netievent_tcpstopchildlisten_t;
 typedef isc__netievent__socket_t isc__netievent_tcpclose_t;
+typedef isc__netievent__socket_t isc__netievent_tcpdnsclose_t;
 typedef isc__netievent__socket_t isc__netievent_startread_t;
 typedef isc__netievent__socket_t isc__netievent_pauseread_t;
 typedef isc__netievent__socket_t isc__netievent_closecb_t;
@@ -644,6 +646,9 @@ isc__nm_tcpdns_close(isc_nmsocket_t *sock);
  * Close a TCPDNS socket.
  */
 
+void
+isc__nm_async_tcpdnsclose(isc__networker_t *worker, isc__netievent_t *ievent0);
+
 #define isc__nm_uverr2result(x) \
 	isc___nm_uverr2result(x, true, __FILE__, __LINE__)
 isc_result_t

lib/isc/netmgr/netmgr.c

@@ -550,6 +550,9 @@ process_queue(isc__networker_t *worker, isc_queue_t *queue) {
 		case netievent_tcpclose:
 			isc__nm_async_tcpclose(worker, ievent);
 			break;
+		case netievent_tcpdnsclose:
+			isc__nm_async_tcpdnsclose(worker, ievent);
+			break;
 		case netievent_closecb:
 			isc__nm_async_closecb(worker, ievent);
 			break;
@@ -675,8 +678,9 @@ nmsocket_cleanup(isc_nmsocket_t *sock, bool dofree) {
 	sock->pquota = NULL;
 
 	if (sock->timer_initialized) {
-		uv_close((uv_handle_t *)&sock->timer, NULL);
 		sock->timer_initialized = false;
+		uv_timer_stop(&sock->timer);
+		uv_close((uv_handle_t *)&sock->timer, NULL);
 	}
 
 	isc_astack_destroy(sock->inactivehandles);
@@ -1022,6 +1026,7 @@ isc_nmhandle_unref(isc_nmhandle_t *handle) {
 	isc_nmsocket_t *sock = NULL;
 	size_t handlenum;
 	bool reuse = false;
+	bool do_close = true;
 	int refs;
 
 	REQUIRE(VALID_NMHANDLE(handle));
@@ -1039,28 +1044,28 @@ isc_nmhandle_unref(isc_nmhandle_t *handle) {
 		handle->doreset(handle->opaque);
 	}
 
-
-
 	/*
 	 * The handle is closed. If the socket has a callback configured
 	 * for that (e.g., to perform cleanup after request processing),
-	 * call it now.
+	 * call it now, or schedule it to run asynchronously.
 	 */
-	bool do_close = true;
 	if (sock->closehandle_cb != NULL) {
 		if (sock->tid == isc_nm_tid()) {
 			sock->closehandle_cb(sock);
 		} else {
-			isc__netievent_closecb_t * event =
+			isc__netievent_closecb_t *event =
 				isc__nm_get_ievent(sock->mgr,
 						   netievent_closecb);
 			isc_nmsocket_attach(sock, &event->sock);
 			isc__nm_enqueue_ievent(&sock->mgr->workers[sock->tid],
 					       (isc__netievent_t *) event);
+
 			/*
-			 * If we do this asynchronously then the async event
-			 * will clean the socket, so clean up the handle from
-			 * socket and exit.
+			 * If we're doing this asynchronously, then the
+			 * async event will take care of closing the
+			 * socket, so we can clean up the handle
+			 * from the socket, but skip calling
+			 * nmsocket_maybe_destory()
 			 */
 			do_close = false;
 		}
@@ -1068,9 +1073,8 @@ isc_nmhandle_unref(isc_nmhandle_t *handle) {
 
 	/*
 	 * We do all of this under lock to avoid races with socket
-	 * destruction.
-	 * We have to do this now otherwise we might race - at this point
-	 * the socket is either unused or attached to event->sock.
+	 * destruction.  We have to do this now, because at this point the
+	 * socket is either unused or still attached to event->sock.
 	 */
 	LOCK(&sock->lock);
 
@@ -1092,15 +1096,8 @@ isc_nmhandle_unref(isc_nmhandle_t *handle) {
 	}
 	UNLOCK(&sock->lock);
 
-	/* Close callback will clean everything up */
-	if (!do_close) {
-		return;
-	}
-
-
-	if (atomic_load(&sock->ah) == 0 &&
-	    !atomic_load(&sock->active) &&
-	    !atomic_load(&sock->destroying))
+	if (do_close && atomic_load(&sock->ah) == 0 &&
+	    !atomic_load(&sock->active) && !atomic_load(&sock->destroying))
 	{
 		nmsocket_maybe_destroy(sock);
 	}

lib/isc/netmgr/tcp.c

@@ -1003,8 +1003,9 @@ tcp_close_direct(isc_nmsocket_t *sock) {
 		}
 	}
 	if (sock->timer_initialized) {
-		uv_close((uv_handle_t *)&sock->timer, timer_close_cb);
 		sock->timer_initialized = false;
+		uv_timer_stop(&sock->timer);
+		uv_close((uv_handle_t *)&sock->timer, timer_close_cb);
 	} else {
 		isc_nmsocket_detach(&sock->server);
 		uv_close(&sock->uv_handle.handle, tcp_close_cb);

lib/isc/netmgr/tcpdns.c

@@ -79,7 +79,6 @@ static void
 timer_close_cb(uv_handle_t *handle) {
 	isc_nmsocket_t *sock = (isc_nmsocket_t *) uv_handle_get_data(handle);
 	INSIST(VALID_NMSOCK(sock));
-	sock->timer_initialized = false;
 	atomic_store(&sock->closed, true);
 	isc_nmsocket_detach(&sock);
 }
@@ -489,10 +488,42 @@ isc__nm_tcpdns_send(isc_nmhandle_t *handle, isc_region_t *region,
 }
 
 
-void
-isc__nm_tcpdns_close(isc_nmsocket_t *sock) {
+static void
+tcpdns_close_direct(isc_nmsocket_t *sock) {
 	if (sock->outer != NULL) {
 		isc_nmsocket_detach(&sock->outer);
 	}
-	uv_close((uv_handle_t *) &sock->timer, timer_close_cb);
+	/* We don't need atomics here, it's all in single network thread */
+	if (sock->timer_initialized) {
+		sock->timer_initialized = false;
+		uv_timer_stop(&sock->timer);
+		uv_close((uv_handle_t *) &sock->timer, timer_close_cb);
+	}
+}
+
+void
+isc__nm_tcpdns_close(isc_nmsocket_t *sock) {
+	REQUIRE(VALID_NMSOCK(sock));
+	REQUIRE(sock->type == isc_nm_tcpdnssocket);
+
+	if (sock->tid == isc_nm_tid()) {
+		tcpdns_close_direct(sock);
+	} else {
+		isc__netievent_tcpdnsclose_t *ievent =
+			isc__nm_get_ievent(sock->mgr, netievent_tcpdnsclose);
+
+		ievent->sock = sock;
+		isc__nm_enqueue_ievent(&sock->mgr->workers[sock->tid],
+				       (isc__netievent_t *) ievent);
+	}
+}
+
+void
+isc__nm_async_tcpdnsclose(isc__networker_t *worker, isc__netievent_t *ievent0) {
+	isc__netievent_tcpdnsclose_t *ievent =
+		(isc__netievent_tcpdnsclose_t *) ievent0;
+
+	REQUIRE(worker->id == ievent->sock->tid);
+
+	tcpdns_close_direct(ievent->sock);
 }